ITR: Knowledge Discovery in Historical Semistructured Data

University Of Maryland, College Park, College Park MD

Investigators

Abstract

The vast majority of data used by scientists, engineers, and decision makers resides in a poorly structured collection of reports, memos, and other documents in a myriad of file formats. The increasing densities and falling prices of storage devices make it practical to store for perpetuity all such data that crosses a scientist's electronic desktop. The resulting Information History has the potential to serve as an intelligent assistant by detecting trends and patterns, suggesting potential collaborators, uncovering relevant documents and data from diverse sources, etc., resulting in dramatic increases in the effectiveness of information use. However, current technology, which focuses on either fully structured or completely unstructured databases, cannot be effectively adapted to extracting knowledge from a large historical semistructured database. The goal of the proposed research is to develop suitable formulations of the knowledge discovery problem for historical semistructured databases and to develop, implement, and evaluate solutions. The research thrusts are: (1) devising knowledge discovery operators for semistructured data and an algebra for them; (2) developing efficient methods for implementing these operators; (3) determining methods to incorporate structure incrementally and flexibly; and (4) incorporating differential processing. A Personal Information History Assistant application serves as a test-bed for this research.